Author(s): Ingrid Keupers; Patrick Willems
Linked Author(s): Patrick Willems
Keywords: Monte Carlo Simulation; Reservoir Model; Sensitivity Analysis; Water Quality
Abstract: The European Water Framework Directive imposes on member states to reach a good water quality status of all their river bodies by 2015. To asses if this goal will be reached and, if not reached, which measures can be undertaken to remediate the situation, an integrated hydrological-hydrodynamic-water quality river model is required that can simulate the concentrations of relevant water quality state variables in the river under given boundary condit ions and for long term rainfall series. Since so many different inputs and processes determine these concentrations, the resulting integrated (physically based) model has an excessively high calculation time. Calibration of the parameters that govern the water quality transformation processes has serious practical difficulties. However, when the detailed model is transformed into a conceptual model by placing non-linear reservoirs in series to represent the river, the computational times can be reduced with a factor 5. 103. This approach is applied to the Grote Laak, a river of 13 km situated in the North East of Belgium. To test the robustness of the conceptual model, and thus asses the validity of this global sensitivity analysis approach, 10 random parameter sets are ran both with the detailed and the conceptual model. Comparison of the conceptual with the detailed model results show that the degree of similarity between both models is such that the conceptual model can be used for time consuming processes such as the calibration of model parameters, scenario runs and long term simulations for statistical processing. The resulting conceptual model is used for sensitivity analysis by means of Monte Carlo simulations to determine the most sensitive parameters of the 23 calibratable parameters that define the water quality transformation processes. More specifically, Latin hypercube sampling was used to select 5000random parameter sets from their probability distribution as determined by expert judgment. The simulated 90th and 50th percentile of dissolved oxygen (DO) ,biological oxygen demand (BOD) ,ammonia (NH4) ,nitrate (NO3) and temperature (Temp) is calculated as this value is used to determine compliance to water quality standards. A multiple linear regress ion is performed on the standardized inputs and outputs of the model to obtain the standardized regression coefficient (SRC) for each of the parameters. The absolute values of these SRC’s allow f or a ranking of the parameters. From this analysis it is clear that the first order BOD decay rate is the most important parameter in the uncertainty of the model result as it ranks in the top three of most influential parameters for both DO, BOD and NH4 at all locations. The third most important parameter in terms of magnitude of sensitivity is the sedimentation rate of BOD. Both parameters indicate the need for a good knowledge on the composition of the BOD in the river and thus could indicate were additional measurement efforts should be spend on. The global sensitivity analysis indicates that from the 23 model parameters, only 9 have a significant contribution to the uncertainty in the model output result. Further calibrate should thus focus on these 9 parameters while keeping the values of the other parameters constant at the recommended value.
Year: 2015